> ## Documentation Index
> Fetch the complete documentation index at: https://sitegpt.ai/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# Chat history

> Read every conversation, reply as a human, and turn missed answers into training.

**Chat History** is the record of everything your chatbot has said. Use it to
check answer quality, reply to escalations, and improve the chatbot with
real data.

## The conversation list

Each conversation shows the visitor (name or anonymous), when it happened,
message count, and status markers for escalation, importance, and feedback.
Filters narrow the list by date, feedback (positive or negative), status,
model, tags, and more.

<Tip>
  The **Downvoted by User** filter is the fastest route to answer quality:
  every downvote is a visitor telling you exactly which answer to fix.
</Tip>

## Inside a conversation

Open a conversation to read the full transcript with sources for each
answer. From here you can:

* **Reply as a human.** Type in the reply box and the visitor sees your
  message in the widget in real time. The AI stays quiet once you take over.
* **Correct an answer.** On a bot message, select **Incorrect? Modify
  bot's answer!** (after a downvote it reads **Downvoted by user! Want to
  modify?**), then write the answer that should have been given. It becomes
  a [custom response](/docs/features/custom-responses) immediately.
* **Mark important** to flag conversations for your team.
* **Resolve** to close out handled conversations.
* **Tag** conversations for later filtering. Conversations started from a
  conversation starter are tagged with that starter automatically.
* **Edit the title** so the list stays scannable.

## Export

Use the download button on the Chat History page to export the current view
as CSV for analysis or record keeping.

## A simple weekly routine

1. Filter by negative feedback and fix each miss: add the missing content or
   an exact custom response.
2. Skim escalated conversations to spot themes the chatbot should learn.
3. Check the unanswered patterns against your
   [conversation starters](/docs/features/conversation-starters); the questions
   visitors type are the starters you should offer.